On Some General Computation Schemes and Hybrid Optimization Techniques used in Learning Processes

摘要

The aim of this presentation is to introduce two general schemes used in learning processing. The first one is a generic reinforcement scheme and the second one a scheme for building SVMs kernels. Both schemes are parameters dependent and the improvement of their computational performances is dependent on the choice of these parameters. In the case of the generic reinforcement scheme the performance is measured in number of iterations in learning process and in the case of SVM kernels in the classification accuracy and cross-validation accuracy obtained during many classification tasks. Different kind of genetic algorithms are used for learning parameters optimization.